Tracking Control Of The De-Oiling System Based On Reinforcement Learning

Oil and water separation facilities are widely used to separate oil and water during offshore oil production and ensure that the oil content of the separated water meets standards for discharge or reinjection into wells.However,the physical model of the de-oiling system is complex,and the violent disturbance of oil-water-gas mixture also brings great challenges to the stability control of the equipment.Meanwhile,the small cavity of the hydrocyclone also brings the constraint of input saturation,which increases the difficulty of the stability control of the de-oiling system.Therefore,this paper studies de-oiling system from modeling,closed-loop system optimization and stability analysis to model-free reinforcement learning optimization algorithm design and other aspects.First of all,the basic knowledge of reinforcement learning required in this paper is introduced,the basic theory of reinforcement learning and common algorithms are discussed,and then combined with the idea of optimal control for general discrete time system optimization problem solving is briefly summarized.Then,considering that the oil-water separation efficiency of the de-oiling system is susceptible to interference input,the interference suppression problem is modeled as the H-infinity tracking control problem of the discrete time de-oiling system,which is transformed into a two-person zero-sum differential game problem of control input and disturbance input.A model-free off-policy reinforcement learning algorithm is proposed.Considering the L₂gain condition to solve the model-based solution,the stability of the system and the stability boundary of the discount factor are analyzed,and a off-policy reinforcement learning algorithm based on state feedback is proposed to solve the H-infinity control problem independent of the system model.The simulation study verifies the effectiveness of the algorithm.Finally,aiming at the saturation problem of the flow control valve of hydrocyclone in oil-water separation facilities,the H-infinity tracking control problem of discrete time de-oiling system under input saturation is further studied.Hyperbolic tangent function is introduced to smoothly fit the saturation state,the cost function is reconstructed,and a off-policy reinforcement learning control algorithm under input saturation is proposed.The Actor-Critic neural network is used to realize the independent learning of the optimal tracking control strategy based on data drive,and the simulation research verifies the effectiveness of the solution.